What a 3 a.m. World Cup kick-off revealed about the Swiss grid.

On the night of 11 to 12 July, a good part of Switzerland set an alarm for three in the morning. The national team was playing Argentina in a World Cup quarter-final, kicking off at 03:00 CEST. We build forecasting systems for the Swiss grid, so we asked the obvious question the next morning: what does a country full of early alarm clocks do to the power system?^[1][2]

The tournament gave us two natural experiments. Switzerland's 2–0 round-of-32 win over Algeria on 3 July kicked off at 05:00 CEST; the 1–3 quarter-final defeat to Argentina on 12 July kicked off at 03:00 CEST, went to extra time and ended around 05:45. For both mornings we compared actual Swiss total load with the day-ahead forecast, then checked the residual control-area imbalance.^[1][2][3][7]

While the country was watching, the grid barely noticed.

The intuitive story is that a nation switching on televisions in the middle of the night should show up as extra consumption. It does not. ENTSO-E publishes actual total load and the day-ahead forecast for the Swiss bidding zone at hourly resolution, and in both match windows the gap between the two stayed within ordinary variation. Modern televisions draw a few tens of watts; against a system running at around six gigawatts, even a couple of million of them disappear into the noise.^[3][4][6]

Televisions are invisible to the grid. A whole country waking up early is not.

After the final whistle, the morning ramp arrived early.

The quarter-final ran long, and the final whistle came around 05:45. Over the next two hours, between 06:00 and 08:00 CEST, actual load exceeded the day-ahead forecast by an average of 1,129 MW, peaking at 1,317 MW at 07:00. Ranked against the same two hours on every calendar day from 1 May through 12 July, that Sunday morning is first out of 73.^[3][4][5]

The explanation is mundane, which is what makes it useful. Nobody consumed more electricity than usual that morning; the country consumed it earlier. A large share of the population was awake two hours before the point a standard Sunday profile assumes, so the ramp that normally builds towards eight o'clock arrived towards six. On an ordinary morning the forecast error is flat, because the day-ahead profile already contains the ramp. A deviation like this only appears when the country wakes off-schedule.^[3][4]

The Algeria morning shows the same shape. In the two hours after that match, from 08:00 to 10:00 CEST, load ran +720 MW above forecast. We read it as supporting evidence rather than proof: it was a Friday, the window overlaps an ordinary working morning, and a few match-free July mornings produced misses of similar size. The quarter-final is the clean exhibit.^[1][3][4]

A gigawatt that mostly never reached the balancing market.

A 1.3 GW forecast miss sounds like it should be a system event. Swissgrid's residual system balance shows why it was not. The two series measure different things: the load comparison captures what the country did against what was scheduled, while the system balance is what remains after balance groups have adjusted their positions and balancing has acted. They cannot be subtracted from one another, but read side by side they show who caught the error.^[4][7][8]

On the quarter-final morning the control area ran short for about two hours, bottoming at −276 MW at 08:00: real and sustained, but a fraction of the gap in the load schedule. The rest was closed upstream, in the intraday adjustments balance groups make as the morning unfolds. The Algeria morning makes the opposite point. The deepest excursion of that night, −185 MW at 05:45, fell in the middle of the match, exactly where the load data shows consumption on forecast. An imbalance chart alone would have told a good story about the wrong thing.^[4][7][8]

The country missed its forecast by a gigawatt, and the market absorbed almost all of it.

What two mornings can and cannot say.

Two events are a hypothesis, not a law, and we hold this one accordingly. A shifted national wake-up is the most plausible reading of the quarter-final morning; weather, calendar quirks and plain forecast error can all contribute, and two observations cannot separate them cleanly. What the two mornings do establish is narrower and more useful: a static day-ahead load profile encodes assumptions about when a country wakes up, and when those assumptions break, they break by hundreds of megawatts for hours at a time.^[3][4][1][2]

Model the alarm clocks, not the match.

Day-ahead profiles are excellent on days that resemble other days. The interesting failures are the predictable ones: the morning after an overnight national-team game, the transition into a holiday week, any event that moves a few million routines by an hour or two. Those are exactly the windows where short-horizon forecasting has to earn its keep, and where signals that track the morning as it actually unfolds are worth more than another refinement of the static profile.^[4][5]